Source: EurekAlert! / INFORMS — 2026-07-06
Summary
A peer-reviewed study of a real-world Geisinger health-system program, published in INFORMS's Manufacturing & Service Operations Management, found that a machine-learning model flagging patients overdue for colorectal cancer screening — paired with nurse-coordinator outreach — measurably increased colonoscopy completion and was associated with a significant drop in two-year mortality.
Key Takeaways
- Flagged patients were 6% more likely to complete colonoscopy within 3 months, and 6.9% more likely within 6 months, versus similar unflagged patients.
- The program was associated with a 6.2 percentage-point reduction in two-year mortality — a 43% relative decrease versus the control group.
- The model used routine EHR inputs (complete blood count, age, sex) to identify elevated-risk, screening-overdue patients, with no exotic data sources required.
- One of relatively few AI-in-care-delivery studies to report hard mortality outcomes rather than process or efficiency metrics alone.